How do massive black holes get their gas?

We use multiscale smoothed particle hydrodynamic simulations to study the inflow of gas from galactic scales (similar to 10 kpc) down to less than or similar to 0.1 pc, at which point the gas begins to resemble a traditional, Keplerian accretion disc. The key ingredients of the simulations are gas, stars, black holes (BHs), self-gravity, star formation and stellar feedback (via a subgrid model); BH feedback is not included. We use similar to 100 simulations to survey a large parameter space of galaxy properties and subgrid models for the interstellar medium physics. We generate initial conditions for our simulations of galactic nuclei (less than or similar to 300 pc) using galaxy-scale simulations, including both major galaxy mergers and isolated bar-(un)stable disc galaxies. For sufficiently gas-rich, disc-dominated systems, we find that a series of gravitational instabilities generates large accretion rates of up to similar to 1-10M(circle dot) yr(-1) on to the BH (i.e. at less than or similar to 0.1 pc); this is comparable to what is needed to fuel the most luminous quasars. The BH accretion rate is highly time variable for a given set of conditions in the galaxy at similar to kpc. At radii of greater than or similar to 10 pc, our simulations resemble the ` bars-within-bars' model of Shlosman et al., but we show that the gas can have a diverse array of morphologies, including spirals, rings, clumps and bars; the duty cycle of these features is modest, complicating attempts to correlate BH accretion with the morphology of gas in galactic nuclei. At similar to 1-10 pc, the gravitational potential becomes dominated by the BH and bar-like modes are no longer present. However, we show that the gas can become unstable to a standing, eccentric disc or a single-armed spiral mode (m = 1), in which the stars and gas precess at different rates, driving the gas to sub-pc scales (again for sufficiently gas-rich, disc-dominated systems). A proper treatment of this mode requires including star formation and the self-gravity of both the stars and gas (which has not been the case in many previous calculations). Our simulations predict a correlation between the BH accretion rate and the star formation rate at different galactic radii. We find that nuclear star formation is more tightly coupled to active galactic nucleus activity than the global star formation rate of a galaxy, but a reasonable correlation remains even for the latter.